[cdob]

Windows 2000 and XP use different cache behaviour.
I guess that's the main difference. I don't have a full explanation.

Uwe describes some XP USB behaviour.
http://www.uwe-siebe...usbstick_e.html

A average USB stick at default settings:
FAT, FAT32: no write cache
NTFS: with write cache

A USB stick is often marked removable.
As knwon a factory tool can flip this setting, e.g. Lexar BootIT.
Or a filter driver can flip this setting. Cfadisk.sys can be integrated at textmode.
http://www.msfn.org/board/index.php?s=&...st&p=818429

Policy 'Optimize for quick removal' is stored in registry.
That's a hardware related setting:

[HKLM\SYSTEM\CurrentControlSet\Enum\USBSTOR\#specific HardwareID#\Device Parameters\Classpnp]
"UserRemovalPolicy"=dword:00000002

Setting can be integrated at migrate.inf or setupreg.hiv.

I wonder:
Does write cache enable a read cache too?

Given a USB stick and write cache active:
What's text mode copy time for FAT, FAT32 and NTFS?

jaclaz, on Jan 30 2009, 04:44 AM, said:

Just for the record, MS released exFAT drivers for XP SP2 or SP3:

Well, how to build a bootable exFAT USB stick at XP?
Do you update mbrbatch.cmd and mkimg.cmd?

format /FS:exFAT creates 32kb cluster at a 2gb USB stick.
At first glance cache behaviour is not changed. Writing small files is slow, no write cache by default.

[jaclaz]

cdob, on Jan 30 2009, 08:25 PM, said:

Well, how to build a bootable exFAT USB stick at XP?
Do you update mbrbatch.cmd and mkimg.cmd?

I have no idea, don't even know if it will EVER be bootable.

I mean, exFAT support needs to be integrated in NTLDR/SETUPLDR.BIN, otherwise there is no way for intial stages of booting, maybe it is possible to "enhance" current "advanced" method "Fake Signature"/"XP Kansas City Shuffle" to support using a "normal filesystem" kicker image and switch during the shuffle the filesystem?

But anyway we'll need to wait for exFAT32 support in grub4dos....

About updating the batches, as soon as I find some time to do that, I hope I will be able to , though if the above speculation is true there would be not much of a point, would it?

jaclaz

[ilko_t]

cdob, on Jan 30 2009, 12:25 PM, said:

I wonder:
Does write cache enable a read cache too?

Maybe it's still all about the write cache? Setup attempts to delete the compressed files while expanding them to target directory. USB devices are write protected, but maybe those attempts for deletion are still speeded up by the enabled write-cache triggered by using NTFS:
http://www.uwe-siebe...usbstick_e.html

Quote

Obviously Windows XP doesn't enable a write cache for USB drives that appear as 'Removable'. The settings 'Optimize for quick removal' or 'Optimize for performance' doesn't seem to make any difference then, except that the latter enables the user to format 'Removable' USB drives with NTFS. But with NTFS Windows XP enables a write cache, writing small files becomes lightning fast.

Seems using NTFS triggers use of write cache even on removable drives.

Removing write-protection gives no speed difference in the tests above, write cache still matters.

Could this be the explanation?

A quick test using this approach (no deleted files during Text mode), may reveal if that's the case:
http://www.msfn.org/...howtopic=119742

I'll have no test rig around for some weeks, any taker?

This post has been edited by ilko_t: 30 January 2009 - 02:16 PM

[cdob]

jaclaz, on Jan 30 2009, 01:35 PM, said:

I mean, exFAT support needs to be integrated in NTLDR/SETUPLDR.BIN, otherwise there is no way for intial stages of booting

Yes, that's true.

jaclaz, on Jan 30 2009, 01:35 PM, said:

maybe it is possible to "enhance" current "advanced" method "Fake Signature"/"XP Kansas City Shuffle" to support using a "normal filesystem" kicker image and switch during the shuffle the filesystem?

Yes, that's possible. I'm chainbooting exFAT at XP already. Did worked at first trial.
BTW: chainboot works at Windows 2000 and XP64 too, should work at all ntldr or setupldr.bin. Most likely at bootmgr too.

jaclaz, on Jan 30 2009, 01:35 PM, said:

But anyway we'll need to wait for exFAT32 support in grub4dos....

Yes, current grub4dos at MBR: dosn't find grldr at exFAT.

@ilko_t
Yes, write cache maybe the main difference.

As for SetupSourceDevice harddisk:
I dislike this because BIOS may change harddisk number. That's not a universal solution.
Maybe use new grub4dos write: boot grub4dos, write txtsetup.sif, set SetupSourceDevice harddisk

[ilko_t]

cdob, on Jan 30 2009, 03:51 PM, said:

@ilko_t
Yes, write cache maybe the main difference.

As for SetupSourceDevice harddisk:
I dislike this because BIOS may change harddisk number. That's not a universal solution.
Maybe use new grub4dos write: boot grub4dos, write txtsetup.sif, set SetupSourceDevice harddisk

Yep, not universal, but should do the trick just to reveal if the write cache makes the difference, since there are no write attempts when using it.

As for the grub4dos write- I doubt, grub4dos uses BIOS disk numbering, boot disk is hd0. At partition screen the same disk is harddisk1, internal one is harddisk0, even though it's not the first in BIOS boot order.
The only semi-logical explanation I found was that disks numbering is in alphabetical order, as per driver name. I.e. disks on ATAPI.SYS get the first numbers, then on DISK.SYS, then SIL3112.SYS for example. Drive letter assignment still depends on BIOS numbering, fixed/removable is taken into account.
Compare:
http://www.msfn.org/board/index.php?showto...st&p=705697
http://www.msfn.org/board/index.php?showto...st&p=647581


USB disks are at DISK.SYS. I tried to rename it to _isk.sys and amend txtsetup.sif accordingly. This failed, disk.sys is hardcoded and used in other drivers, I did not go further.

Thanks for your input, as always you trigger a little brainstorm.

[BigDaddy]

http://www.uwe-siebe...usbstick_e.html
this is what I have been looking for. So I can finally map my network share as D:.

[jaclaz]

ilko_t, on Jan 31 2009, 09:06 AM, said:

USB disks are at DISK.SYS. I tried to rename it to _isk.sys and amend txtsetup.sif accordingly. This failed, disk.sys is hardcoded and used in other drivers, I did not go further.

Probably also setup.hiv and .inf files need to be amended...

You can try renaming other drivers, i.e. ATAPI.SYS->ZTAPI.SYS, just for the sake of checking?

I don't think it will work, as an ATA/IDE drive will be both under ATAPI.SYS and DISK.SYS.....

jaclaz

[coords]

Just thought that I'd mention that M$ has now made exFAT32 available for WinXP. It can be downloaded here -> http://support.microsoft.com/kb/955704

[BigDaddy]

as long as this update will not be automatically delivered through WU the majority of WinXP users will not have support for exFAT and thus my USB drive becomes much less portable if exFAT is used.

This post has been edited by BigDaddy: 12 February 2009 - 10:57 AM

[pointertovoid]

The answer to the speed issue on Usb sticks between W2k, Xp, Fat, Ntfs seems well explained here:
http://www.uwe-siebe...usbstick_e.html under "Cache or not"

[EDIT: oops, sorry, already mentioned in a post above]

The whole site is very interesting:
http://www.uwe-sieber.de/english.html

This post has been edited by pointertovoid: 19 February 2009 - 07:44 AM

[jaclaz]

Just for the record, a comparison test with different filesystems on different sticks:
http://www.testfreaks.com/blog/information...-ntfs-vs-exfat/

Is it possible that Vista has re-added some speed to FAT32?

jaclaz

[ilko_t]

A few tests on Windows 7 x64.

Buffalo RUF2-R2G USB stick, Windows 7 x64, xcopy-ing I386 folder only, from XP SP3 from a folder on the internal hard disk, 5878 small files, 375MB total. Default cluster size for 2GB partition, tests repeated several times with format before each.



NTFS - 572.22 seconds
FAT16 - 169.17 seconds
FAT32 - 171.19 seconds

Quite slow Apacer 4GB stick:



NTFS - 696.48
FAT16 - 818.19
FAT32 - 903.17

Optimize for performance/removal doesn't seem to make any sugnificant difference in each case.